A5029206786- Handwritten Text Recognition Techniques
- Advanced Image and Video Retrieval Techniques
- Advanced Vision and Imaging
- Advanced Image Processing Techniques
- Industrial Vision Systems and Defect Detection
- Digital Media Forensic Detection
- Visual Attention and Saliency Detection
- Image Processing and 3D Reconstruction
- Image Retrieval and Classification Techniques
- Advanced Data Compression Techniques
- Smart Grid Energy Management
- Color perception and design
- Gaze Tracking and Assistive Technology
- Water Quality Monitoring Technologies
- Image and Signal Denoising Methods
- Indoor and Outdoor Localization Technologies
- Vehicle License Plate Recognition
- Robotics and Sensor-Based Localization
- Video Coding and Compression Technologies
- Identification and Quantification in Food
- Natural Language Processing Techniques
- Advanced Image Fusion Techniques
- Video Surveillance and Tracking Methods
- Video Analysis and Summarization
- Distributed and Parallel Computing Systems
Ryukoku University
2022-2023
Tohoku University
2012-2022
High Energy Accelerator Research Organization
2019
Advanced Science Research Center
2019
Osaka University
2019
Tohoku University Hospital
2017
École Nationale des Greffes
2010-2016
Tokyo University of Agriculture and Technology
1998
We propose a GAN-based image compression method working at extremely low bitrates below 0.1bpp. Most existing learned methods suffer from blur bitrates. Although GAN can help to reconstruct sharp images, there are two drawbacks. First, makes training unstable. Second, the reconstructions often contain unpleasing noise or artifacts. To address both of drawbacks, our adopts two-stage and network interpolation. The is effective stabilize training. Moreover, interpolation utilizes models in...
We propose an end-to-end consistently trainable text detection method based on the Faster R-CNN. The original R-CNN is CNN for fast and accurate object detection. By considering characteristics of texts, a novel architecture that make use its ability proposed. Although generates region interests (RoIs) by proposal network (RPN) using feature map last convolutional layer, proposed RoIs multiple RPNs maps layers. This uses multiresolution to detect texts various sizes simultaneously. To...
Feature extraction is essential for chemical property estimation of molecules using machine learning. Recently, graph neural networks have attracted attention feature from molecules. However, existing methods focus only on specific structural information, such as node relationship. In this paper, we propose a novel convolutional network that performs with simultaneously considering multiple structures. Specifically, paths specialized in node, edge, and three-dimensional Moreover, an...
Scene text erasing, which replaces regions with reasonable content in natural images, has drawn significant attention the computer vision community recent years. There are two potential subtasks scene erasing: detection and image inpainting. Both require considerable data to achieve better performance; however, lack of a large-scale real-world scene-text removal dataset does not allow existing methods realize their potential. To compensate for pairwise data, we made use synthetic after...
This paper addresses the automatic generation of a typographic font from subset characters. Specifically, we use to extrapolate additional Consequently, obtain complete containing number characters sufficient for daily use. The automated Japanese fonts is in high demand because requires over 1,000 Unfortunately, professional typographers create most fonts, resulting significant financial and time investments generation. proposed method can be great aid creation designers do not need majority...
Recently, attention has surged concerning intelligent sensors using text detection. However, there are challenges in detecting small texts. To solve this problem, we propose a novel detection CNN (convolutional neural network) architecture sensitive to scale. We extract multi-resolution feature maps multi-stage convolution layers that have been employed prevent losing information and maintain the size. In addition, developed considering receptive field size generate proposal stages. The...
In Japan, blue and chub mackerels are often caught simultaneously, their market prices different. Humans need to sort them manually, which requires heavy labor. The demand for automatic sorting machines is increasing. aim of this paper develop an machine mackerels, a challenging task. There two required functions. First, it needs localization on conveyor belt so that can be transported destinations. Second, species classification needed, but difficult due similar appearance among the...
With the widespread use of smartphones, there is a surging demand for localization in indoor environments. The main challenges are requirement special equipment (e.g., map database and Wi-Fi access points) error accumulation localization. In this paper, we propose novel collaborative positioning method to reduce accumulation. Estimated positions corrected using collaborator’s when an encounter detected by communication based on Bluetooth Low Energy (BLE). addition, obtained taking photos...
One of the major causes traffic accidents according to statistical report on in Japan is disregard lights by drivers. It would be useful if driving support systems could detect and recognize give appropriate information Although many studies intelligent transportation have been conducted, detection using images remains a difficult problem. This because are very small as compared other objects there similar road environment. In addition, pixel colors easily over-saturated, which renders light...
This paper proposes a method for automatic discrimination of two mackerel species: Scomber japonicus (chub mackerel) and australasicus (blue mackerel). Because S. has much higher market price than australasicus, the species must be properly sorted before shipment, but their similar appearance makes difficult. These can effectively distinguished using ratio base length between dorsal fin’s first ninth spines to fork length. However, manual measurement this is time-consuming reduces fish...
When a disaster occurs, video communication is an effective way to disseminate large quantities of important information. However, coding standards such as High Efficiency Video Coding (HEVC) compress entire videos, whatever the contents are; at low bit rates, quality significant objects deteriorates. In this paper, object-based method proposed address problem. The extracts on basis visual saliency and temporal correlation between frames. Subsequently, we execute pre-processing which...
This paper describes the design and construction of stainless steel tank JSNS2 detector. The leakage was examined using water gas after construction. new sealing technique with liquid gasket developed, its capability evaluated quantitatively. result shows over 5 times better performance than tolerance level leakage. acceleration measurement during transportation adequate robustness. These tests prove that is feasible to be used in experiment.
In this work, we propose a method for estimating the user's gazing point in environment using images taken by an eye tracker and omnidirectional camera. The proposed estimates positon mapping obtained camera image. However, matching image is difficult because distorted equirectangular projection. Therefore, location to with considering distortion. Specifically, repeats conversion results.
JSNS$^2$ (J-PARC Sterile Neutrino Search at J-PARC Spallation Neutron Source) is an experiment that searching for sterile neutrinos via the observation of $\bar{\nu}_{\mu} \rightarrow \bar{\nu}_e$ appearance oscillations using with muon decay-at-rest. For this search, rejecting cosmic-ray-induced neutron events by Pulse Shape Discrimination (PSD) essential because detector located above ground, on third floor building. We have achieved 95$\%$ rejection while keeping 90$\%$ signal,...
There exist distributed processing environments composed of many heterogeneous computers. It is required to schedule parallel processes in an appropriate manner. For the scheduling, prediction execution load a process effective exploit resources environments. We propose long-term methods with references properties and runtime predictions. Since method different according situation, we module selection select state changing CPU using neural network. also discuss about implementation system,...
Lossy image compression causes a loss of texture, especially at low bitrate. To mitigate this problem, we propose novel method that utilizes reference-based super-resolution model. We use two models and self texture transfer The encode decode whole input selected reference patches. patches are small but compressed with high quality. model transfers the into similar regions in image. experimental results show our can reconstruct accurate by transferring
Distributed power generation, using renewable energy, has been attracting attention to cope with global environment issues; a microgrid is promising configuration for distributed generation. To augment the stability and efficiency of microgrid, an intelligent control, which considers restrictions characteristics each unit, indispensable. It can be achieved by constructing efficient operation schedule plant in depending on energy demand, predicting passive The scheduling regarded as...
Various indoor navigation methods have been developed recently, but digitalized data of map is not always available. Therefore, an framework using image information board has proposed. In this method, the process to extract regions from necessary be done by hands beforehand, and estimate passageway important because its used in matching. However, method discrimination very heuristic, which intended for a specific type floor maps. paper, we propose semi-automatic with simple user's operation....
This study proposes a lightweight solution to estimate affine parameters in motion compensation. Most of the current approaches start with an initial approximation based on standard estimation, which only estimates translation parameters. From there, iterative methods are used find best parameters, but they require significant amount time. The proposed method aims speed up process two ways, first, skip evaluating prediction when it is likely bring no encoding efficiency benefit, and second,...
Improvement of character recognition technology brings us various applications for mobile camera. However, many low-resolution and poor-quality images exist due to the performance camera or influence environment, existing methods are not good at recognizing those characters. Therefore, we develop a system ultra-low resolution less than 20*20 pixels. The proposed consists three phases: increased training data with generative learning method, creating deblurred high-resolution image Wiener...
The development of information technology has added many conveniences to our lives. On the other hand, however, we have deal with various kinds information, which can be a difficult task for elderly people or those who are not familiar devices. A recognize each person’s activity and providing appropriate support based on that could useful such people. In this paper, propose novel fine-grained recognition method user systems focuses identifying text at is gazing, idea content related user. It...
Pansharpening is a method applied for the generation of high-spatial-resolution multi-spectral (MS) images using panchromatic (PAN) and images. A common challenge in pansharpening to reduce spectral distortion caused by increasing resolution. In this paper, we propose reducing based on intensity–hue–saturation (IHS) targeting satellite The IHS improves resolution an RGB image replacing intensity low-resolution with that high-resolution PAN image. characteristics MS are different, difference...
Pansharpening (PS) is a process used to generate high-resolution multispectral (MS) images from high-spatial-resolution panchromatic (PAN) and high-spectral-resolution images. In this paper, we propose method for pansharpening by focusing on compressed sensing (CS) technique. The spectral reproducibility of the CS technique high due its image reproducibility, but reproduced blurry. Although methods complementing incomplete reproduction have been proposed, it known that existing may cause...